Hi David,
ANCOVA is just regression, with a bit less information. So try to
think of this more broadly.
You have a slope interaction, that means you can't use ANCOVA in the
conventional sense, but ANCOVA is just regression, Don't throw the
baby out with the bathwater.
Here's your problem. What you do with ANCOVA is evaluate the
difference between the two groups, controlling for some other measure.
You evaluate the difference when the control variable is equal to
zero. You then hope that the slopes are parallel, so that you can say
that the difference IS the difference.
You don't have parallel slopes, and zero is a really bad place to
evaluate the difference, as no one scores anything close to zero on
the control variables.
One approach is to look at regions of significance - that is, to say
"at what values of age or WASI (or whatever) do I find significant
differences. http://people.ku.edu/~preacher/interact/
The easier thing to do is just to look at the graphs. To make life
easier, I would center the control variables at the mean, so the
difference between the groups (and the significance of the difference
between the groups) is evaluated at the mean of the control variables.
Jeremy
On 29 April 2011 09:09, David Hambrook <[log in to unmask]> wrote:
> Hi there
> I'm having some trouble interpreting the results of an ANCOVA I have run for
> my DClinPsy thesis project. I wonder if anyone might be able to help. My
> Design is:
> DV: Emotional Intelligence Score
> IV: Group (Anorexia vs. Controls)
> Covariates: Age, IQ, Anxiety, Agreeableness
>
> It looks like there is an interaction between 3 of my covariates (Age, IQ,
> Anxiety) and the IV (see plots attached). If I run an ANCOVA and include
> interaction effects between each of these covariates and group they are
> significant. I think this means that I have violated the assumption of
> homogeneity of regression slopes?
> Does this mean that I can't use ANCOVA with my data? If I can still use it,
> could you advise (in layman's terms) how to do it (using SPSS), how I can
> justify this given the assumption violation, and how to report it? If I
> can't use ANCOVA, are there any alternatives?
> I would be SOOO grateful for any help with this, I've been going round in
> circles and keep coming back very confused :(
> Thanks in advance
> Dave
> PS - I'm happy to send my data to anyone who needs to see it before they can
> help.
> ________________________________
> David Hambrook
> Trainee Clinical Psychologist
> Royal Holloway
> University of London
>
>
>
>
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|